A package for analysis of rare particle decays with machine-learning algorithms
Project description
raredecay
This package consists of several tools for the event selection of particle decays, mostly built on machine learning techniques. It contains:
a data-container holding data, weights, labels and more and implemented root-to-python data conversion as well as plots and KFold-data splitting
reweighting tools from the hep_ml-repository wrapped in a KFolding structure and with metrics to evaluate the reweighting quality
classifier optimization tools for hyper-parameters as well as feature selection involving a backward-elimination
an output handler which makes it easy to add text as well as figures into your code and automatically save them to a file
… and more
HowTo examples
To get an idea of the package, have a look at the howto notebooks: HTML version or the IPython Notebooks
Minimal example
Want to test whether your reweighting did overfit? Use train_similar:
from raredecay.data import HEPDataStorage
from raredecay.score import train_similar
mc_data = HEPDataStorage(df, weights=*pd.Series weights*, target=0)
real_data = HEPDataStorage(df, weights=*pd.Series weights*, target=1)
score = train_similar(mc_data, real_data, old_mc_weights=1 *or whatever weights the mc had before*)Getting started right now
If you want it the easy, fast way, have a look at the Ready-to-use scripts. All you need to do is to have a look at every “TODO” task and probably change them. Then you can run the script without the need of coding at all.
Documentation and API
The API as well as the documentation: Documentation
Setup and installation
Depending on which functionality you want to use, you may consider different installations.
Everything (reweighting incl. scoring, machine learning, ROOT bindings)
Follow the instructions for each dependency separately and then install raredecay via
pip install raredecay[all]
Machine learning, advanced scores
First install the following version of REP (the -U can be omitted, but is recommended to have the newest dependencies, on the other hand may crashes REPs reproducibility):
pip install -U https://github.com/yandex/rep/archive/stratifiedkfold.zip
Then install the package via
pip install -U raredecay[ml]
Reweighting (without scoring)
To install the newest version of hep_ml containing the loss-regularization (recommended, but optional).
pip install -U git+https://github.com/arogozhnikov/hep_ml.git
Then, install the raredecay package (without ROOT-support) via
pip install raredecay
To make sure you can convert ROOT-NTuples, use
pip install raredecay[root] # *use raredecay\[root\] in a zsh-console*
As it is a young package still under developement, it may receive regular updates and improvements and it is probably a good idea to regularly download the newest package.
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
File details
Details for the file raredecay-2.0.1.tar.gz.
File metadata
- Download URL: raredecay-2.0.1.tar.gz
- Upload date:
- Size: 94.0 kB
- Tags: Source
- Uploaded using Trusted Publishing? No
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 | 99ee5555e7a6feeb7e8b8428d25b4beb17ac684f6bbac16e04df96928204772d |
|
| MD5 | e1822eef5802f506f3bcd00aea1236b2 |
|
| BLAKE2b-256 | 4fd81335ac6ee4b25a26f61a27f62beab28804e9f9b12dfdd989899205822dfc |
File details
Details for the file raredecay-2.0.1-py3-none-any.whl.
File metadata
- Download URL: raredecay-2.0.1-py3-none-any.whl
- Upload date:
- Size: 112.0 kB
- Tags: Python 3
- Uploaded using Trusted Publishing? No
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 | 9432d5548f8fceca8c0a0cb844a54e462a63676b0831a1f64cd1e05a695c000e |
|
| MD5 | b4bfa65f6e3eac3f2e2c6d7fb3adb531 |
|
| BLAKE2b-256 | 9a05f73060a903e8bbe97f550b56b2d4ada75f13c2ae941065aaa38e625e3558 |
